(Jackendoff 1996a) is tothinkthatnotallsyntacticconfigurationsare inherentlymeaningful—thattherelationbetween
form and meaning is oftenmoreflexiblethan in constructions like those we have seen here. For instance, sections 5.8
and 5.9 argued that the range of meanings associated with transitiveand ditransitive verb constructions and withN-of-
NPconstructions is too great for the meaning to be attributed to the construction itself. Rather, all the semantic
relations come from the head verb or noun, and the construction is merely a form to befilled.
The view of constructions as lexical items, however, suggests an alternative possibility. If there are “defective”
constructions like the resultativethat lack phonology, why couldn't therebe“doubly defective”constructions that lack
bothphonology and semantics? Such constructions would be phrase structure rules! For example, (16), the expansion
of VP into V plus NP plus PP/AP/Prt, could be such a construction. It could be clipped together with other lexical
“treelets”(for example the expansion of S into NP plus VP and the expansion of NP into Det plus N) to form
standard phrase-structure trees.
This manner of viewing tree construction begins to blur the distinction between lexical items and what have
traditionally been regarded as rules of grammar. It reflects the same spirit as our treatment of regular morphology not
in terms of rules that add affixes, but rather as free combination of lexically stored parts. In this approach, the only
“rule of grammar”is UNIFY PIECES, and all the pieces are stored in a common format that permits unification.^87
Ordinary words, which encode word-sized linked units of phonology, syntax, and semantics, are on the other end of a
scale (or better, at an opposite corner of a multidimensional space) from phrase-structure rules, which encode phrase-
sized pieces of syntax without any associated phonology or semantics. But we can attest every step on the scale in
between them—and most of these steps are in fact types of unit that have raised difficulties for standard
syntactocentric architectures. It is too soon to tell how this progra mof consolidation will work out, but it sets an
interesting agenda for future research.^88
180 ARCHITECTURAL FOUNDATIONS
(^87) In a sense this proposal thus begins to converge with the Minimalist Program, which creates structure by means of the single operation Merge, and with Chomsky's view
(1995) that all differences among languages are encoded in the lexicon. A major difference, of course, is that the present approach has no counterpartto the Minimalist
Program's operation Move—not to mention the crucial architectural differences discussed in Ch. 5. A much closer parallel, though, is with unification-based gram-mars,
especially HPSG and Construction Grammar. See section 6.11 for some comparison.
(^88) An important problem for the constructional approach that has not to my knowledge been solved satisfactorily is how constructional idioms compose with each other.
Exampleslike(i),a middleconstructionbased ona resultative,and (ii),a relativeclause basedona comparativecor-relative,aresufficientlyremotefromeverydayexperience
that one would guess that their constructional properties must be composed online.(i) That pot doesn't cook black so easily.
(ii) This is a book that the more you study, the more confused you get.
In the traditionalderivationalformalism, of course, constructions like thesearise from the successive applicationof movementrules. But successiveapplicationis not availablein
a constraint-based formalism.